2
$\begingroup$

My data follows a sigmoidal function of the form
$$y=asym/(1+e^{(xmid-x)/scale)})$$
I have taken the function from the SSLogis function in R.
My supervisor and I think that there is a second variable that influences the asymptote and scale of the function.

So I'm stuck with the questions:
1. Is it at all possible to do a multiple regression based on this function?
If yes, 2. Is it possible to transform the sigmoidal to a linear function and use it in a multiple linear regression?
Or 3. Is there a way to do a multiple non-linear regression?

Any help is really welcome.

$\endgroup$
1
  • 1
    $\begingroup$ library(nlme); help("gnls") or if you have repeated measures help("nlme"). $\endgroup$
    – Roland
    Apr 7, 2016 at 10:49

1 Answer 1

3
$\begingroup$

Certainly you can have multiple predictors in nonlinear least squares.

Exactly how it works depends on how, precisely, you want the second variable to enter.

You could make an equation that (at a fixed asymptote) has a linear predictor by simple reparameterization:

Let $\alpha$ be your "asym" and let $\beta_0=-$xmid/scale. Let $\beta_1$ be 1/scale and let $x_1$ be your $x$.
Let $x_2$ be the new predictor (independent variable, explanatory variable), and let $\beta_2$ be its coefficient.

Then you could write

$$E(y)=\frac{\alpha}{(1+e^{-(\beta_0+\beta_1x_1+\beta_2x_2)})}\quad^*$$

and then $\beta_2=0$ would correspond to your original model.

$^*$ (note that the LHS is $E(y)$ not $y$; the model for $y$ includes an error term)

Whether this is what you need depends on how you see the second variable coming in.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.